Event hub

AGI Odds: How Markets Price the Race to General AI

"AGI odds" refers to the prices on prediction markets that estimate when, or whether, artificial general intelligence will arrive — and they are among the most-watched and most-argued numbers in tech. Platforms like Polymarket and Kalshi list contracts such as "OpenAI announces it has achieved AGI before 2027," and forecasting sites like Metaculus run longer-horizon questions on when the first general AI system will be publicly announced. The appeal is obvious: a single price seems to compress thousands of expert opinions into one probability. The problem is just as obvious once you look closely — there is no agreed definition of AGI, so what each market is actually asking varies widely, and the resolution criteria do most of the work. This page explains what these markets ask, how they are framed and resolved, what news moves the odds, and why the definitional ambiguity means any AGI price should be read with unusual care. None of this is financial advice.

No live markets on this are trading right now — browse the live feed for what the crowd is pricing today.

What "AGI odds" markets actually ask

There is no single AGI market. Instead there is a family of contracts, and they ask very different questions despite sharing the same three letters.

The most common type is a milestone-announcement market tied to a specific lab and a deadline. On Polymarket, for example, "OpenAI announces it has achieved AGI before 2027?" resolves to Yes only if OpenAI or an official representative states that it has created AGI by the end of 2026. Kalshi runs a parallel contract on whether OpenAI announces AGI by December 31, 2026. Note what these markets actually price: not whether AGI exists in some philosophical sense, but whether a particular company publicly declares it within a window. A model could be transformative without anyone calling it "AGI," or a company could use the term loosely — the contract only tracks the announcement.

A second type is a dated-forecast question, most associated with Metaculus rather than a money market. Its long-running question on "when the first general AI system will be devised, tested, and publicly announced" attaches a concrete checklist — passing adversarial Turing-style tests, scoring well on broad exams, and completing certain tasks — so the resolution is far more demanding than a press release. As of early 2026 that community forecast pointed to the early 2030s, with a wide range, after forecasters pushed their timelines later through 2025 and 2026.

A third type prices the competitive race rather than the finish line: "which company has the best AI model" by a given date. These are not AGI markets at all, but they sit in the same category and often move on the same news.

How these markets resolve

Resolution is where AGI markets live or die, because the term itself carries no fixed meaning. A money market on Polymarket or Kalshi resolves on an observable event — almost always a public announcement — rather than on whether a model is "truly" general. For the OpenAI contracts, the primary resolution source is official information from OpenAI or its representatives, with a consensus of credible reporting used as a backstop. If the named company has not made the announcement by the deadline, the market resolves No.

This design is deliberate. An observable, reportable trigger can be adjudicated; a contested abstraction like "human-level intelligence" cannot. The trade-off is that the market answers a narrower question than the headline suggests. "Will OpenAI announce AGI by 2027" depends partly on capability and partly on corporate and legal incentives to use the word at all.

Those incentives are not hypothetical. Reporting on the Microsoft–OpenAI partnership described an internal, contractual definition that reportedly tied AGI to a profit threshold — figures around $100 billion in profits have been cited — because the label carried real consequences for the two companies' rights. Whether or not that specific framing still governs the relationship, it shows how far a working definition of AGI can sit from any benchmark a researcher would recognize. When you read an AGI price, read the market's Rules tab first: the resolution source and the exact deadline matter more than the word "AGI" in the title.

The core problem: AGI has no agreed definition

Every AGI odds figure rests on a definition that the field has never settled, and this is the single most important caveat. Some use AGI to mean a system that matches or exceeds humans across most economically valuable cognitive work. Others mean a system that can learn any new task as efficiently as a person. The Microsoft–OpenAI context reportedly used a financial threshold. Each definition implies a different market, a different resolution date, and a different probability.

Benchmarks illustrate the gap rather than closing it. The ARC-AGI test, created by François Chollet in 2019 to measure the ability to generalize to genuinely novel tasks, became a focal point when OpenAI's o3 model reportedly scored roughly 87% on the first version (ARC-AGI-1) in late 2024 — a result Chollet called a significant step-function increase. But the harder ARC-AGI-2, released in 2025, saw top scores fall sharply, and an interactive ARC-AGI-3, introduced in 2026, reportedly left frontier models near zero while humans solved its tasks. Strong performance on one benchmark plainly does not equal AGI, and the benchmarks themselves keep moving as systems improve.

The practical consequence: two markets can show very different AGI odds and both be internally consistent, because they are not asking the same thing. A near-term lab-announcement contract can look frothy while a strict, checklist-based forecast points years further out. Treat any AGI probability as conditional on its specific resolution criteria, not as a reading of "how close we are" in the abstract.

What moves the odds

AGI and AI-leadership markets react to a fairly predictable set of inputs, even if the magnitude of each move is hard to call in advance.

Model releases are the biggest driver. A new frontier model, especially one with a step-change on reasoning or agentic tasks, can swing both the milestone markets and the "best model" race quickly. In June 2026, for instance, a frontier release reshaped the "best AI model" market, with the crowd concentrating around a single leader and tens of millions of dollars trading on the question. Benchmark results move odds in the same way — a surprising jump (or a flat result) on tests like ARC-AGI, broad exams, or coding and agent evaluations is read as evidence about the trajectory.

Lab and executive statements matter because the milestone contracts resolve on announcements. Public comments from figures like Dario Amodei and Demis Hassabis are watched closely; reported timelines have ranged from "a few years" to "five to ten years," and notably, several prominent forecasters lengthened their AGI timelines during 2025–2026 rather than shortening them. Independent forecasts such as the AI 2027 scenario, and later revisions pushing key thresholds toward the early 2030s, feed the same debate.

Finally, liquidity and attention move prices in ways unrelated to capability. Monthly volume on AI markets has fluctuated and at points declined through 2026, and thin markets can produce noisy odds. A small number of large traders can move a quiet contract, so a price is only as informative as the volume and the time left to resolution behind it.

Reading AGI odds without overreading them

A prediction-market price is a money-weighted estimate of probability — useful precisely because participants have something at stake — but on a contested, long-horizon question like AGI, several limits apply at once.

First, the price answers the resolution criteria, not the headline. "40% chance of AGI by year X" almost always means "40% chance a named entity makes a specified announcement by a specified date," which is narrower and more announcement-driven than the phrase suggests.

Second, crowd-implied probabilities can be wrong, and they are wrong more often on novel, hard-to-verify questions than on, say, an election with a fixed counting date. There is no historical base rate for "AGI by 2027," so the market is aggregating opinion about an unprecedented event, not pricing a well-understood frequency.

Third, time and liquidity shape the number. Far-dated contracts carry large uncertainty and can be moved by a single release; thin markets can drift on low volume. The honest way to use AGI odds is as one input among many — a snapshot of where informed money sits today, alongside benchmark results, lab statements, and independent forecasts — rather than as a settled probability. We report these prices as a signal of what the crowd believes; we do not treat them as predictions you should act on, and nothing here is financial advice.